datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
textvqa
Large-scale Multi-modality Models Evaluation Suite
Accelerating the development of large-scale multi-modality models (LMMs) with lmms-eval
🏠 Homepage | 📚 Documentation | 🤗 Huggingface Datasets
This Dataset
This is a formatted version of TextVQA. It is used in our lmms-eval pipeline to allow for one-click evaluations of large multi-modality models.
@inproceedings{singh2019towards,
title={Towards vqa models that can read},
author={Singh, Amanpreet and… See the full description on the dataset page: https://huggingface.co/datasets/lmms-lab-encoder/textvqa.hle_text_only
Humanity's Last Exam - (Text only)
🌐 Website | 📄 Paper | GitHub
Center for AI Safety & Scale AI
Humanity's Last Exam (HLE) is a multi-modal benchmark at the frontier of human knowledge, designed to be the final closed-ended academic benchmark of its kind with broad subject coverage. Humanity's Last Exam consists of 3,000 questions across dozens of subjects, including mathematics, humanities, and the natural sciences. HLE is developed globally by subject-matter experts and… See the full description on the dataset page: https://huggingface.co/datasets/macabdul9/hle_text_only.dummy_image_text_data
Dataset Card for "dummy_image_text_data"
More Information needed
Textground4MTextGround4M: A Prompt-Aligned Dataset for Layout-Aware Text Rendering
TextGround4M is a large-scale dataset for prompt-grounded, layout-aware text rendering in text-to-image (T2I) generation, introduced in our AAAI 2026 paper.
Dataset Summary
TextGround4M contains 4.1 million prompt-image pairs, each annotated with:
A natural language caption where all rendered text spans are explicitly quoted
Span-level bounding boxes linking each quoted… See the full description on the dataset page: https://huggingface.co/datasets/CSU-JPG/Textground4M.TextAtlas5M
TextAtlas5M
This dataset is a training set for TextAtlas.
Paper: https://huggingface.co/papers/2502.07870
(All the data in this repo is uploaded :>)
Dataset subsets
Subsets in this dataset are CleanTextSynth, PPT2Details, PPT2Structured,LongWordsSubset-A,LongWordsSubset-M,Cover Book,Paper2Text,TextVisionBlend,StyledTextSynth and TextScenesHQ. The dataset features are as follows:
Dataset Features
image (img): The GT image.
annotation (string): The input prompt… See the full description on the dataset page: https://huggingface.co/datasets/CSU-JPG/TextAtlas5M.hle_text_onlyTextEdit
TextEdit: A High-Quality, Multi-Scenario Text Editing Benchmark for Generation Models
Danni Yang,
Sitao Chen,
Changyao Tian
If you find our work helpful, please give us a ⭐ or cite our paper. See the InternVL-U technical report appendix for more details.
🎉 News
[2026/03/06] TextEdit benchmark released.
[2026/03/06] Evaluation code and initial baselines released.
[2026/03/06] Leaderboard updated with latest models.
📖 Introduction… See the full description on the dataset page: https://huggingface.co/datasets/opencompass/TextEdit.flickr_1k_test_image_text_retrieval
Flickr30k (1K test set)
Original paper: From image descriptions to visual denotations: New similarity metrics for semantic inference over event descriptions
Homepage: https://shannon.cs.illinois.edu/DenotationGraph/
1K test set split from: http://cs.stanford.edu/people/karpathy/deepimagesent/caption_datasets.zip
Bibtex:
@article{young2014image,
title={From image descriptions to visual denotations: New similarity metrics for semantic inference over event descriptions}… See the full description on the dataset page: https://huggingface.co/datasets/nlphuji/flickr_1k_test_image_text_retrieval.image-text_medieval-scripts_xiv-xv-xvi
Dataset Card for image-text_medieval-scripts_xiv-xv-xvi
This dataset was created using pagexml-hf converter from Transkribus PageXML data.
Dataset Summary
This dataset contains 548322 samples across 1 split(s).
Geographical scope: BelgiumPeriod: 1350-1550Languages: FlemishType of document: ProtocolProvenance: State Archives in Leuven
Projects Included
Itinera Nova
Parts of Charters from Königsfelden
SAL7304_full
SAL7305_full
SAL7306_full
SAL7307
SAL7307_full… See the full description on the dataset page: https://huggingface.co/datasets/dh-unibe/image-text_medieval-scripts_xiv-xv-xvi.TextPecker-1.5M
TextPecker-1.5M: A Dataset for Training and evaluating TextPecker
This repository contains the TextPecker-1.5M dataset, a new benchmark proposed in the paper "TextPecker: Rewarding Structural Anomaly Quantification for Enhancing Visual Text Rendering".
Code and Project Page
The official implementation and project details for the TextPecker and TextPecker-1.5M dataset can be found on the GitHub repository:
https://github.com/CIawevy/TextPecker
Sample Usage
You… See the full description on the dataset page: https://huggingface.co/datasets/CIawevy/TextPecker-1.5M.Total-Text-DatasetTotal Text Dataset.
It consists of 1555 images with more than 3 different text orientations: Horizontal, Multi-Oriented, and Curved, one of a kind.
Original github repo; https://github.com/cs-chan/Total-Text-Dataset
Forked repo; https://github.com/yunusserhat/Total-Text-Dataset
Describable-Textures-Dataset
Dataset Card for Describable Textures Dataset
This is a FiftyOne dataset with 5640 samples.
Installation
If you haven't already, install FiftyOne:
pip install -U fiftyone
Usage
import fiftyone as fo
import fiftyone.utils.huggingface as fouh
# Load the dataset
# Note: other available arguments include 'max_samples', etc
dataset = fouh.load_from_hub("Voxel51/Describable-Textures-Dataset")
# Launch the App
session = fo.launch_app(dataset)… See the full description on the dataset page: https://huggingface.co/datasets/Voxel51/Describable-Textures-Dataset.Describable-Textures-Dataset-DTD
Not sure about the license.
Source: https://www.robots.ox.ac.uk/~vgg/data/dtd/
Describable Textures Dataset (DTD)
The Describable Textures Dataset (DTD) is an evolving collection of textural images in the wild, annotated with a series of human-centric attributes, inspired by the perceptual properties of textures. This data is made available to the computer vision community for research purposes.
Download… See the full description on the dataset page: https://huggingface.co/datasets/cansa/Describable-Textures-Dataset-DTD.mscoco_2014_5k_test_image_text_retrieval
MSCOCO (5K test set)
Original paper: Microsoft COCO: Common Objects in Context
Homepage: https://cocodataset.org/#home
5K test set split from: http://cs.stanford.edu/people/karpathy/deepimagesent/caption_datasets.zip
Bibtex:
@inproceedings{lin2014microsoft,
title={Microsoft coco: Common objects in context},
author={Lin, Tsung-Yi and Maire, Michael and Belongie, Serge and Hays, James and Perona, Pietro and Ramanan, Deva and Doll{\'a}r, Piotr and Zitnick, C Lawrence}… See the full description on the dataset page: https://huggingface.co/datasets/nlphuji/mscoco_2014_5k_test_image_text_retrieval.TextCaps
Large-scale Multi-modality Models Evaluation Suite
Accelerating the development of large-scale multi-modality models (LMMs) with lmms-eval
🏠 Homepage | 📚 Documentation | 🤗 Huggingface Datasets
This Dataset
This is a formatted version of TextCaps. It is used in our lmms-eval pipeline to allow for one-click evaluations of large multi-modality models.
@inproceedings{sidorov2019textcaps,
title={TextCaps: a Dataset for Image Captioningwith Reading… See the full description on the dataset page: https://huggingface.co/datasets/lmms-lab-encoder/TextCaps.hle_text_onlyGradients_Gradients_and_Text_Full_Logic_CaptionsTextRich
Dataset Composition
This dataset is a multi-domain bennchmark for detecting AI-generated text-rich images from GPT-Image-2.
The dataset consists of two complementary subsets:
Fake subset:
Images generated by GPT-Image-2 using carefully designed prompts. The prompts are constructed to cover diverse domains and layouts while avoiding reproduction of specific real-world images.
Real subset:
Images sampled from six publicly available datasets. These images are selected as… See the full description on the dataset page: https://huggingface.co/datasets/Shuyiww/TextRich.text-dataset-tiny-code-script-py-format
USED of tahamajs/medicine_ds_persian for .parquet file
USED of Alijafarixcs2/persian-it-llama2-2k for .parquet file
USED of Abirate/english_quotes for .jsonl file
NEW FILES (05/12/2025)
NEW FILES (12/26/2025)
NEW FILES (02/15/2026)
Multimodal-Dataset-Image_Text_Table_TimeSeries-for-Financial-Time-Series-ForecastingThe sp500stock_data_description.csv file provides detailed information on the existence of four modalities (text, image, time series, and table) for 4,213 S&P 500 stocks.
The hs300stock_data_description.csv file provides detailed information on the existence of four modalities (text, image, time series, and table) for 858 HS 300 stocks.
If you find our research helpful, please cite our paper:
@article{xu2025finmultitime,
title={FinMultiTime: A Four-Modal Bilingual Dataset for… See the full description on the dataset page: https://huggingface.co/datasets/Y123-wed/Multimodal-Dataset-Image_Text_Table_TimeSeries-for-Financial-Time-Series-Forecasting.COCO-Text
Dataset Card for "COCO-Text"
More Information needed
Total-Text-Dataset
Dataset Card for Total-Text-Dataset
The Total-Text consists of 1555 images with more than 3 different text orientations: Horizontal, Multi-Oriented, and Curved
This is a FiftyOne dataset with 1555 samples.
Installation
If you haven't already, install FiftyOne:
pip install -U fiftyone
Usage
import fiftyone as fo
import fiftyone.utils.huggingface as fouh
# Load the dataset
# Note: other available arguments include 'max_samples', etc
dataset =… See the full description on the dataset page: https://huggingface.co/datasets/Voxel51/Total-Text-Dataset.text-2-video-human-preferences
Rapidata Video Generation Preference Dataset
This dataset was collected in ~12 hours using the Rapidata Python API, accessible to anyone and ideal for large scale data annotation.
The data collected in this dataset informs our text-2-video model benchmark. We just started so currently only two models are represented in this set:
Sora
Hunyouan
Pika 2.0
Runway ML Alpha
Luma Ray 2
Explore our latest model rankings on our website.
If you get value from this dataset and would… See the full description on the dataset page: https://huggingface.co/datasets/Rapidata/text-2-video-human-preferences.Multimodal-Dataset-Image_Text_Table_TimeSeries-for-Financial-Time-Series-ForecastingThe sp500stock_data_description.csv file provides detailed information on the existence of four modalities (text, image, time series, and table) for 4,213 S&P 500 stocks.
The hs300stock_data_description.csv file provides detailed information on the existence of four modalities (text, image, time series, and table) for 858 HS 300 stocks.
If you find our research helpful, please cite our paper:
@article{xu2025finmultitime,
title={FinMultiTime: A Four-Modal Bilingual Dataset for… See the full description on the dataset page: https://huggingface.co/datasets/Wenyan0110/Multimodal-Dataset-Image_Text_Table_TimeSeries-for-Financial-Time-Series-Forecasting.text-2-video-human-preferences-seedance-1-pro
Rapidata Video Generation Seedance 1 Pro Human Preference
In this dataset, ~60k human responses from ~20k human annotators were collected to evaluate Seedance 1 Pro video generation model on our benchmark. This dataset was collected in roughtly 30 min using the Rapidata Python API, accessible to anyone and ideal for large scale data annotation.
Explore our latest model rankings on our website.
If you get value from this dataset and would like to see more in the future, please… See the full description on the dataset page: https://huggingface.co/datasets/Rapidata/text-2-video-human-preferences-seedance-1-pro.t-e-x-t-re-c-a-ptextlatent_zebra_thinkmorph_armAB
Text-Latent (Arm A) vs All-Latent (Arm B) — Zebra-CoT + ThinkMorph
35638 samples/arm, 18 categories. Schema = ULVR/williamium style (sample_id, category, source_dataset,
question, answer, input_image, intermediate_image_N, num_intermediate_steps, messages_json).
armA_text_latent: real decoded text CoT + latent visual blocks (intermediate_image_1..3).
armB_render_latent: reasoning text RENDERED to images, all-latent baseline (intermediate_image_1..17).
messages_json = full Monet… See the full description on the dataset page: https://huggingface.co/datasets/RuoliuYang/textlatent_zebra_thinkmorph_armAB.text-2-video-human-preferences-veo3
Rapidata Video Generation Veo 3 Human Preference
In this dataset, ~46k human responses from ~20k human annotators were collected to evaluate Veo3 video generation model on our benchmark. This dataset was collected in roughly 35 minutes using the Rapidata Python API, accessible to anyone and ideal for large scale data annotation.
Explore our latest model rankings on our website.
If you get value from this dataset and would like to see more in the future, please consider liking… See the full description on the dataset page: https://huggingface.co/datasets/Rapidata/text-2-video-human-preferences-veo3.dhivehi-image-text
Dhivehi Image-Text Dataset
A dataset of Dhivehi (Maldivian) image-text pairs for machine learning and computer vision tasks.
Dataset Statistics
Total number of batches: 10
Total images across all batches: 394,212
Average images per batch: ~39,421
Split ratios:
Training: 80%
Validation: 10%
Test: 10%
Batch Details
Batch
Total Images
Train
Validation
Test
dv01-01
39659
31727
3966
3966
dv01-02
38989
31191
3899
3899
dv01-03
39360
31488
3936
3936… See the full description on the dataset page: https://huggingface.co/datasets/alakxender/dhivehi-image-text.ImageNet-100-Text